A Physics-Informed, Global-in-Time Neural Particle Method for the Spatially Homogeneous Landau Equation

TL;DR

Numerical experiments on analytical benchmarks, including the two- and three-dimensional BKW solutions, as well as reference-free configurations, demonstrate stable transport, preservation of macroscopic invariants, and competitive or improved accuracy compared with time-stepping score-based particle and blob methods while using significantly fewer particles.

Abstract

We propose a physics-informed neural particle method (PINN--PM) for the spatially homogeneous Landau equation. The method adopts a Lagrangian interacting-particle formulation and jointly parameterizes the time-dependent score and the characteristic flow map with neural networks. Instead of advancing particles through explicit time stepping, the Landau dynamics is enforced via a continuous-time residual defined along particle trajectories. This design removes time-discretization error and yields a mesh-free solver that can be queried at arbitrary times without sequential integration. We establish a rigorous stability analysis in an $L^2_v$ framework. The deviation between learned and exact characteristics is controlled by three interpretable sources: (i) score approximation error, (ii) empirical particle approximation error, and (iii) the physics residual of the neural flow. This trajectory estimate propagates to density reconstruction, where we derive an $L^2_v$ error bound for kernel density estimators combining classical bias--variance terms with a trajectory-induced contribution. Using Hyvarinen's identity, we further relate the oracle score-matching gap to the $L^2_v$ score error and show that the empirical loss concentrates at the Monte Carlo rate, yielding computable a posteriori accuracy certificates. Numerical experiments on analytical benchmarks, including the two- and three-dimensional BKW solutions, as well as reference-free configurations, demonstrate stable transport, preservation of macroscopic invariants, and competitive or improved accuracy compared with time-stepping score-based particle and blob methods while using significantly fewer particles.

Figures (25)

• Figure 1: Conceptual architecture of PINN--PM. Left: global-in-time trajectory network $\Phi_\xi$ (shared parameters across particles). Right: global-in-time score network $s_\theta$. Both networks are trained jointly through a physics residual enforcing the Landau characteristics and an implicit score-matching loss. The framework removes explicit time stepping and enables direct querying at arbitrary times.
• Figure 2: Global-in-time architecture of PINN--PM. Separate velocity and time embeddings are processed by neural blocks and fused to produce the continuous-time trajectory $v^{(i)}_t$. The architecture enables mesh-free querying in time without explicit time stepping.
• Figure 3: Error-analysis flow for PINN--PM. Score/measure/residual errors enter the drift mismatch and yield a Grönwall-type trajectory bound; synchronous coupling closes the estimate; ISM and KDE provide downstream score/density certificates.
• Figure 4: BKW-2D: one-dimensional density slices. Density slices along $(x,0)$ (top) and $(0,y)$ (bottom) at $t\in\{1,2.5,5\}$ Curves compare PINN--PM, SBP, Blob, and the analytical solution. Density is reconstructed via KDE with bandwidth $\varepsilon=0.15$.
• Figure 5: BKW-2D: particle transport snapshots. Particle positions at $t\in\{1,2.5,5\}$. The plot shows (i) a reference Euler-integrated particle evolution, (ii) the PINN--PM predicted particle locations, and (iii) score directions (learned vs. analytic). The close overlap indicates that the learned trajectory network captures the characteristic transport.

Theorems & Definitions (22)

• Remark 1: Domain and Boundary Conditions
• Remark 2
• Lemma 1: Properties of the exact and approximate drifts
• proof
• Theorem 1: Trajectory error bound with PINN residual
• proof
• Lemma 2: Hyvärinen oracle identity
• proof
• Lemma 3: Empirical--oracle gap
• proof